PhD Oral Exam - Maksym Perepichka, Computer Science

When studying for a doctoral degree (PhD), candidates submit a thesis that provides a critical review of the current state of knowledge of the thesis subject as well as the student’s own contributions to the subject. The distinguishing criterion of doctoral graduate research is a significant and original contribution to knowledge.

Once accepted, the candidate presents the thesis orally. This oral exam is open to the public.

Abstract

The simulation of garments for virtual game characters is a widely studied field. Traditionally, video games have relied on skinning to animate character garments, with limited quality. Physics-based simulation can be used to generate quality draped garment deformations, but the associated computational cost renders it unsuitable for real-time applications. Neural networks have recently been utilized as general function approximators to simulate garment behavior at a fraction of the cost of traditional physics simulation. Two classes of neural simulators exist: graph-based and pose-based. Graph-based neural networks offer high quality results and generalizability, but remain computationally expensive. Pose-based methods are less robust, but are much faster, allowing for their execution in real time. This thesis explores four avenues of research to improve pose based neural cloth simulators: rigid garment deformation, loose garment simulation, material dependence, and distillation of graph-based methods. We find that many of these features of graph-based methods can be effectively supported by pose-based methods, while leaving the door open for future research to fully close the gap between the two paradigms.